This document describes how to configure callable tools (MCP services, Catalog functions, and model services) for an Agent, enabling the Agent to call external capabilities.
Overview
In DataBuddy, an Agent uses external capabilities through the Tools mechanism:
Tools: Call encapsulated capabilities such as MCP services, Catalog functions, model services, SQL query engines, Volumes, and Jobs.
The standard protocol for tools is MCP (Model Context Protocol). For details, see MCP Overview. This document focuses on how to attach tools to an Agent. Prerequisites
An Agent has been created. For details, see Creating and Managing Agents.
The current account has access to the MCP services and Catalog functions to be used.
To add an MCP service, you must have the MCP management permission for the corresponding workspace.
Tool Configuration (High-Code)
Calling Tools in High-Code Agents
A high-code Agent calls MCP tools in code through the DataBuddy SDK, without requiring developers to handle user authentication separately:
Purpose | SDK call direction |
Obtain a large model instance | Create a model client directly by model ID (such as wedata-gpt-5-2). |
Call MCP tools | List/call MCP services visible under the current account through the SDK. |
Record traces | Automatically write to the corresponding experiment through the MLflow SDK. |
For specific method signatures, refer to the SDK documentation. Development process:
1. Open the Agent code template in Studio.
2. In your code, import the DataBuddy SDK as needed, and use Skills such as discover_tools to search for available MCPs by platform.
3. Explicitly declare the set of callable tools in the Agent orchestration logic.
4. After deployment, the Agent can call tools on demand at runtime.
Tip
You can automatically search for available MCP services on the platform through AI-assisted tools in Studio, avoiding manual lookup.
Other Configurations (User Authorization and Billing)
Item | Description |
User authentication | When calling MCP, Agent uses its own service account, so developers do not need to handle OAuth authentication in code. |
Call logs | All tool calls and model calls are automatically written to the MLflow experiment associated with the Agent. For details, see Agent Debugging and Testing. |
Billing | Billing is not available currently. A separate notice will be provided after the official commercial launch. |
FAQs
Q: How Are Authentication Details (API Keys and Tokens) for MCP Tools Managed?
The authentication information for MCP tools is configured in the "Add External MCP Tool" process in MCP Overview and stored in the workspace's secure credential store. When an Agent calls the tools, the platform automatically injects the authentication information, so developers do not need to handle API keys directly in their code. References
Creating and Managing Agents